AI Revolution: ChatGPT and iMessage Unite  🔗 **Source:** [AI Revolution: ChatGPT and iMessage Unite](https://share.snipd.com/episode/e03a812b-76b7-49ff-9e79-34319732c21f) ## Highlights - **ChatGPT Can Fully Access Your iMessage Locally** - ChatGPT's iMessage plugin can read, draft, search, and send messages when allowed on-device. - Jaeden Schafer demoed drafting follow-ups and deep-searching message history across ChatGPT, Codex, and ChatGPT for Work. Transcript: Jaeden Schafer About what's going on with chat GPT. So like I mentioned, it just got access to iMessage. And this is actually via a plugin. So they originally launched the plugin for chat GPT. And then they also have led it and they've also added it to chat to work and codex so basically all of the open ai all of the all the open ai tools can access your iMessage if you allow it of Course you have to go and allow it they can sort it can read it can draft and it can send iMessages now there's a bunch of places that my mind goes to immediately one of them is that there's A ton of tools one in particular called SendBlue, where like for salespeople and cold outreach, they use these different specific tools that will allow it to basically look like you're Texting from an iPhone. And they do this so that when someone gets a text on an iPhone, if it looks like it's coming from an iMessage, they have a higher perceived trust versus if it's a green text message from Any other sort of SMS sender. So to me, immediately, my mind went there. It's because that's an interesting industry and an interesting business. And now you essentially could automate that let chat GPT have access to your iMessage. And you could probably let it have access to your businesses, quote unquote, iMessages that you would have on your computer. And you could get, you know, verified blue bubble text messages for free. So that's like one interesting benefit but i think a lot of people are concerned about the privacy element i think even they were kind of warning opening i warned against enabling persistent Approval which is what i usually give chad chupiti and claude when it's working on something i look you get unlimited approval just do whatever you need and it you know it still asks me Like if it has to copy and paste a password it's gonna make me do that. But for a lot of the tasks, I don't want it to ask me every five seconds, like the direction of the task it's taking. But apparently, they're warning in this case, it's not great, because then it's going to remove your final chance to review anything that you said. Now, if you're using this to like organize your life and to reply to your grandma, you probably want to, you know, review what's being sent here. But if this is something that you're using for maybe work or maybe kind of like sales, and you're, you know, doing high volumes, 100% this is something that I would use opening, I told Bloomberg that the plugin runs locally on your device, so it doesn't create an index of all of someone's messages, all of that is happening on your device. A lot of people are worried about the privacy concerns of this. It's like I mentioned, going to work with Codex and ChatGPT work. And they had this kind of launch demo. They basically showed ChatGPT drafting some follow up messages based on the prior days conversations and searching deep message history. So honestly, that's kind of useful if it's going to go down a deep rabbit hole and get some information from a chat you had a long time ago, ChatGPT can go and do the research for you, which Is pretty useful. ([Time 0:02:30](https://share.snipd.com/snip/15612738-fd29-42f8-b9cd-2dbf95daf19f)) - **iMessage Access Enables Blue Bubble Outreach** - Granting ChatGPT iMessage access could enable free 'blue bubble' style outreach previously done by tools like SendBlue. - Jaeden links this to sales/cold outreach advantages from appearing as native iPhone texts. Transcript: Jaeden Schafer And they do this so that when someone gets a text on an iPhone, if it looks like it's coming from an iMessage, they have a higher perceived trust versus if it's a green text message from Any other sort of SMS sender. So to me, immediately, my mind went there. It's because that's an interesting industry and an interesting business. And now you essentially could automate that let chat GPT have access to your iMessage. And you could probably let it have access to your businesses, quote unquote, iMessages that you would have on your computer. And you could get, you know, verified blue bubble text messages for free. ([Time 0:03:12](https://share.snipd.com/snip/7ab60154-f5ca-46bc-908e-1db79ecd8e9e)) - **Avoid Persistent Approval For Automated iMessages** - Avoid enabling persistent approval for the iMessage plugin unless you trust automated sends without final review. - Jaeden warns persistent approval removes your final chance to review outgoing messages. Transcript: Jaeden Schafer So that's like one interesting benefit but i think a lot of people are concerned about the privacy element i think even they were kind of warning opening i warned against enabling persistent Approval which is what i usually give chad chupiti and claude when it's working on something i look you get unlimited approval just do whatever you need and it you know it still asks me Like if it has to copy and paste a password it's gonna make me do that. But for a lot of the tasks, I don't want it to ask me every five seconds, like the direction of the task it's taking. But apparently, they're warning in this case, it's not great, because then it's going to remove your final chance to review anything that you said. Now, if you're using this to like organize your life and to reply to your grandma, you probably want to, you know, review what's being sent here. But if this is something that you're using for maybe work or maybe kind of like sales, and you're, you know, doing high volumes, 100% this is something that I would use opening, I told Bloomberg that the plugin runs locally on your device, so it doesn't create an index of all of someone's messages, all of that is happening on your device. ([Time 0:03:43](https://share.snipd.com/snip/5a79ddee-aeb1-42c3-9130-ce57e7ce267f)) - **Smart Glasses Detector Apps Scan Bluetooth Nearby** - Detector apps scan nearby Bluetooth to alert you when smart glasses (Meta Ray-Bans, Snapchat Spectacles) might be nearby. - Jaeden notes apps like Nearby Glasses, Zuck Off, and Anti-Zuck reached high downloads and charts quickly. Transcript: Jaeden Schafer And you can get on iOS and Android and it is at about 100,000 downloads on the Play Store. This just came out in February. There's another one called Zuck Off that launched it's free on iPhone. It just launched and it is a $10 a year perpetual perpetual like price for it. But essentially what these things do is they're actually scanning the Bluetooth of all the surrounding devices. And anytime a device that is, you know, a meta Ray-Bans, or even if it's something like the Snapchat Spectacles, if there's any of these smart glasses that pop up on their radar, if they Hit them with the Bluetooth, they're going to notify you and let you know that, look, there's probably some meta Ray-Bans around you that could be recording you. So I think a lot of people enjoy are happy with that meta reinstalled and then they also unreleased some facial recognition systems on about 50 million phones in june they actually deleted It because wired and eff kind of exposed it and talked about it so they went and removed that but at defcon 2026 they actually banned smart glasses completely. There's a bunch of schools, courts, restaurants, entertainment venues that have also banned smart glasses. There's another app called Anti-Zuck Smart Glasses Scanner that got to very close to the top of the iPhone charts in July. It was $3. They're kind of all in that same ballpark. So Ives Jean Mard, who described himself on GitHub as a no pro developer, warned that detecting Bluetooth LE devices might sometimes just not work as expected and false positives are Likely. So to be fair, these aren't always accurate. ([Time 0:05:44](https://share.snipd.com/snip/7c17d0d7-e365-405c-a2aa-6ed63c15278b)) - **Detector Apps Risk False Positives From Bluetooth** - Bluetooth LE detection can be unreliable and produce false positives, making detector apps imperfect. - Jaeden quotes a GitHub developer warning that false positives are likely with Bluetooth scans. Transcript: Jaeden Schafer Positives are likely. So to be fair, these aren't always accurate. And I think that might be kind of worrisome for some people if you're laying in bed one night, and all of a sudden your phone beeps that there's smart glasses and you look around and there's No one near you, that would be kind of an awkward, you know, false positive. ([Time 0:07:12](https://share.snipd.com/snip/e1da8e65-2ddd-443f-babb-67e3f7e01563)) - **Big AI Labs Petition For Pacing Automated Research** - About 1,100 AI lab employees from big firms asked the US government to help pace automated AI research. - Jaeden frames this as coming from large companies that may be trying to shape governance in ways favoring incumbents. Transcript: Jaeden Schafer There are a hundred and a hundred thousand, there is 1,100 AI lab employees that are asking the US government to help pace automated AI research. There's a bunch of people from OpenAI, Anthropic, Google, and Meta. Essentially, what they're concerned about in particular is that automated AI research could accelerate beyond human control without coordinated governance. Now, it seems kind of like a perhaps a regulatory capture approach because this is OpenAI, Anthropic, and Google, and Meta. These are the biggest companies that are going to have a lot of budget to spend on this kind of stuff. They have the most money to spend, you know, like hundreds of billions of dollars, or in the case of Anthropic, a billion dollars a month just on, you know, data centers for data processing And for running their models. So some of these companies have an insane amount of money. And so when they're going to the government and saying, hey, look, we're worried that like our training is going to accelerate beyond human control, you need to put a bunch of like, governance In place. I'm a little bit pessimistic, because it feels like they're trying to pull up the ladder behind them. These aren't small companies asking for this. These are the big companies that have big budgets that, you know, theoretically, they're the ones that could mistakenly have their training model, perhaps run and accelerate beyond What they're capable of doing. There's also, I think, Microsoft Mistral and Thinking Machines that signed the pacing the frontier statement. All of the people were basically asking the government to back an internal effort to build tools that can deliberately slow automated AI research. OpenAI had their CRO, Mark Chen and chief scientist, Jacob Paco, and also their co-founder, John Shulman, and some others sign it. Anthropic had their co-founder, Jack Clark, sign it. So there's a lot of people that are working on this right now. The statement follows last week's incident where you had OpenAI escape in the sandbox, hacking hugging faces, or maybe the week before. So obviously, this is something that people are concerned about. They're like, look, if the AI model's out there hacking things, we don't want it to progressively get better and better. In my opinion, you can always unplug the machine. I know people probably hate that statement. But at the end of the day, I think that I don't always love a lot of this regulation that comes out. Perhaps some of it's necessary, but it feels like there's a bunch of questions that are not being answered in this. It doesn't specify how pacing would be verified across jurisdictions, how China-based labs would be brought into any coordination framework. And the spoiler alert is they wouldn't be. So all of these US companies are going to decide that, hey, there's this kind of like coordinated framework where we're not going to let the AI model train itself to make itself better And faster. And do you think the Chinese open source models are going to care? Absolutely not. They're going to just run with that. So they give the end of the day, this probably isn't going to pass. And it probably isn't super viable because of that. ([Time 0:07:31](https://share.snipd.com/snip/a7c5f6e8-524c-47d3-a551-4d2fe5c46915)) - **Data Labeling Firms Are Exploding With Demand** - MicroOne grew from $100M to $500M gross run-rate in eight months, reflecting massive demand for labeled training data. - Jaeden explains data labeling and packaged datasets can have 80–90% gross margins and high contractor retention issues. Transcript: Jaeden Schafer Micro One has just hit $500 million in run rate as AI training data demand is surging. This is a company that's only four years old. It's a data labeling company and they grew five fold in eight months. So they're right behind Merkur who is at $2 billion and Handshake is at $1 billion. Both of all three of those companies are in the same market. Micro One's gross annual run rate grew from about $100 million to $500 million over the last eight months. That is a 5x growth rate. That is absolutely insane. The net run rate lands between about $150 million and $200 million after 60% to 70% contract contractor retention. Merkur hit $2 billion in gross annualized revenue this summer and Handshake reached $1 billion earlier this year. So what's interesting to me is what these companies are actually selling off the shelf data sets sold to multiple customers are about 80 to 90% gross margin. And MicroOne raised their series A at a $500 million valuation last September. And they might have also recently raised again at a higher market, although it wasn't exactly disclosed. This is actually an area that I'm working in. I'm specifically working with podcasters and people that have video interviews. I have a bunch of companies that I work with where we license data to them and they'll, you know, purchase data sets of essentially people talking in high quality audio and video to help Train general purpose. It's not making video or voice clones of people, but it's helping train AI models on how to talk, how conversations happen. They sell this to a number of different AI companies. And it's actually a very lucrative, especially if someone has a big data set of a bunch of recordings or interviews that they've done over the years. It's a really great way to make money from something that you already have. I think a lot of people are realizing this and it is a massive growing industry. Now, a lot of these companies here aren't just selling the data, but they're also labeling it. They're also processing the data. There's a ton of work that has to go into actually making this data usable for AI companies when they're training their models. And so there's a lot that goes into it. ([Time 0:10:17](https://share.snipd.com/snip/e12d5d22-7238-43be-8f56-2d96df0e3407)) - **Every Built An Editor Clone From 30,000 Edits** - Every trained an editing agent on 30,000 edits to clone editor-in-chief Kate Lee's judgment and doubled staff from ~15 to ~30. - Jaeden recounts Every's backtesting system that checks edits against Kate Lee's historical approvals. Transcript: Jaeden Schafer A company called Every has built an AI clone of their editor in chief. And they did that using 30, they trained it on 30,000 edits. Their CEO is Dan Shipper. And he says that they have a 30 person team. And they were able to double their head count while they automated all of their copy desk. They built an agent, Kate Lee was their editor in chief, and they built an agent that cloned all of Kate Lee's editing, which is very wild. The CEO said, we have an editor in chief, Kate Lee, who's fantastic, who I've been trying to automate out of a job for years in an extremely benevolent way. I mean, I think they're benevolent because she has a lot of work to do. But every single collected one of her 300 or of her 30,000 historical edits from being an editor in chief, they were used to train an internal copy editing agent that back tests against Her past work. So it will go and edit something and then it will go check, you know, based off her past work, would this be approved? Would this work? The company doubled, like I mentioned, from about 15 people to about 30 people over the last year. They're running six software products, and they also have a daily newsletter. ([Time 0:12:23](https://share.snipd.com/snip/d27794fc-099e-462f-b6e9-43982fe0da20))